A Molecular Retro-Synthesis Method, System, Storage Medium and Terminal Based on Quantum Graph Convolution

Abstract

The present invention discloses a molecular reverse synthesis method, system, storage medium and terminal based on quantum graph convolution, belonging to the field of drug analysis. The structural information and chemical properties of the target product are represented in the form of a molecular undirected graph; each graph node and connection edge are mapped to a unique vector by using one-hot encoding; then the molecular graph node and connection edge information data are quantum encoded; a quantum graph convolutional neural network is constructed based on the encoded quantum graph, and the position of the predicted reaction center is output to obtain a synthon graph; a graph convolutional policy network is used to traverse the synthon graph, identify the atoms that need to be replaced or added and map them into an action vector, and decode the action vector to obtain the representation of the reactant. The present invention realizes the translation process from the product graph to the reactant graph, ensures the usability of the final reactant, solves the problems of large computational difficulty and unsatisfactory synthesis route prediction results in the current molecular reverse synthesis task, and provides a technical reference for future drug research and development.

Description

Technical Field

[0001] The present invention relates to the field of pharmaceutical analysis, and particularly to a molecular retrosynthesis method, system, storage medium and terminal based on quantum graph convolution. Background Art

[0002] Traditional molecular retrosynthesis requires researchers to have rich chemical knowledge and experimental experience in order to effectively design synthesis schemes, select appropriate reagents and reaction conditions, and solve problems that may occur during the experiment. This requires researchers to have an in-depth understanding and mastery of aspects such as chemical reaction mechanisms, structure-activity relationships, reaction kinetics, and analytical testing techniques. In addition, in the experiments of molecular retrosynthesis, factors such as the purity of reactants, the feasibility of synthesis routes, the yield and purity of synthesis products also need to be considered, and the reaction process needs to be carefully monitored and controlled. Due to the combined influence of these factors, traditional molecular retrosynthesis often requires a large amount of time and resources, and various difficulties and challenges are often encountered in practice. Therefore, the use of new technologies such as machine learning can provide a faster, more efficient and accurate method for molecular retrosynthesis, thus greatly reducing the experimental cost and risk.

[0003] In the field of molecular retrosynthesis, common machine learning methods include recursive neural networks (RNNs), Transformer models, convolutional neural networks, etc. Among them, RNNs and Transformer models are common methods for molecular retrosynthesis based on sequence matching. However, since these methods cannot effectively constrain the availability of reactants, the prediction results of synthesis routes may be unsatisfactory, and manual intervention is still required for the selection of specific reactants and the design of synthesis routes. In contrast, convolutional neural networks can extract and represent the local structure of molecules, but due to their spatial limitations, they cannot fully consider the global structure information of molecules, resulting in possible errors in prediction results. Graph neural networks are a framework for learning graph-structured data using deep learning, which can comprehensively consider the topological and spatial structure information of molecules and use operations such as graph convolution to process and model molecular graphs. Therefore, they have good application prospects in the field of molecular retrosynthesis.

[0004] However, due to the complexity and diversity of chemical reaction mechanisms, a large number of intermediate products and transition states are involved in the reaction process, which increases the difficulty of reaction pathway modeling and the prediction results of synthesis routes are unsatisfactory. At the same time, the computational tasks of a large amount of drug molecule graph data pose a huge challenge to traditional computers. Summary of the Invention

[0005] The object of the present invention is to overcome the problems of high difficulty in large-scale molecular graph calculation and unsatisfactory prediction results of synthetic routes commonly existing in current molecular retrosynthesis tasks, and provides a molecular retrosynthesis method, system, storage medium and terminal based on quantum graph convolution.

[0006] The object of the present invention is achieved by the following technical solutions:

[0007] In the first aspect, a molecular retrosynthesis method based on quantum graph convolution is provided, including the following steps:

[0008] S1. Represent the structural information and chemical properties of the target product in the form of a molecular undirected graph, where the atoms of the target product are used as graph nodes and the chemical bonds between atoms are used as connecting edges;

[0009] S2. Map each graph node and connecting edge into a unique vector in the way of one-hot encoding; then perform quantum encoding on the molecular graph node and connecting edge information data to convert it into a corresponding quantum graph;

[0010] S3. Construct a quantum graph convolutional neural network based on the encoded quantum graph, output the position of the predicted reaction center, disconnect the chemical bond where the reaction center is located, and then separate to obtain a synthon graph;

[0011] S4. Traverse the synthon graph using a graph convolutional policy network, identify the atoms that need to be replaced or added, map them into an action vector, and decode the action vector to obtain the representation of the reactant;

[0012] S5. Repeat steps S1 - S4, continuously update the reactant until the synthon graph is converted into the final reactant.

[0013] As a preferred option, in a molecular retrosynthesis method based on quantum graph convolution, the quantum encoding of the molecular graph node and connecting edge information data includes:

[0014] Encoding and converting the input vector into a superposition quantum state using the angle encoding method, and the conversion formula is as follows:

[0015]

[0016] where θ represents the rotation angle of the quantum rotation gate, and x i represents the vector obtained by one-hot encoding.

[0017] As a preferred option, in a molecular retrosynthesis method based on quantum graph convolution, the quantum graph convolutional neural network includes a quantum graph convolutional layer, a quantum pooling layer and a quantum fully connected layer.

[0018] As a preferred option, a molecular retrosynthesis method based on quantum graph convolution, the step S3 specifically includes:

[0019] S31. Apply Hadamard gates and phase rotation gate operations to the input quantum circuit to perform a convolution operation on the encoded quantum state, thereby extracting the features of the quantum graph and updating the state to achieve information transmission;

[0020] S32. Sum each tensor output by the quantum graph convolution, use CNOT gates to control whether the summation result of each tensor is retained, and use SWAP gates to combine all the retained tensors, thereby obtaining a pooling result;

[0021] S33. Flatten the output of the quantum pooling layer into a vector, perform information processing through a quantum neural network layer, and output the prediction result of the reaction center.

[0022] As a preferred option, a molecular retrosynthesis method based on quantum graph convolution, the information processing through the quantum neural network layer and outputting the prediction result of the reaction center includes:

[0023] The quantum graph convolution neural network outputs a reaction center probability distribution vector, and selects the edge of the reaction center probability distribution vector with the highest score as the reaction center.

[0024] As a preferred option, a molecular retrosynthesis method based on quantum graph convolution, before traversing the synthon graph using the graph convolution policy network, further includes:

[0025] Use a graph convolution neural network to extract the features of nodes and edges according to the input synthon graph, and generate feature vectors of nodes and edges.

[0026] As a preferred option, a molecular retrosynthesis method based on quantum graph convolution, the identifying atoms that need to be replaced or added and mapping them into an action vector includes:

[0027] Generate an action vector based on the feature vectors of nodes and edges for updating the synthon graph.

[0028] In a second aspect, a molecular retrosynthesis system based on quantum graph convolution is provided, and the system includes:

[0029] A molecular undirected graph construction module configured to represent the structural information and chemical properties of the target product in the form of a molecular undirected graph, where the atoms of the target product are used as graph nodes and the chemical bonds between the atoms are used as connecting edges;

[0030] The quantum state encoding module is configured to map each graph node and connection edge into a unique vector in the form of one-hot encoding; then perform quantum encoding on the molecular graph node and connection edge information data to convert it into a corresponding quantum graph;

[0031] The reaction center recognition module is configured to construct a quantum graph convolutional neural network based on the encoded quantum graph, output the predicted position of the reaction center, break the chemical bond where the reaction center is located, and thus separate to obtain a synthon graph;

[0032] The synthon-reactant conversion module is configured to traverse the synthon graph using a graph convolutional policy network, identify the atoms that need to be replaced or added and map them into an action vector. After repeating multiple times, decode the action vector to obtain the representation of the reactant.

[0033] In a third aspect, there is provided a computer storage medium, on which computer instructions are stored, and when the computer instructions run, they execute the relevant content in any one of the above-mentioned molecular reverse synthesis methods based on quantum graph convolution.

[0034] In a fourth aspect, there is provided a terminal, including a memory and a processor. The memory stores computer instructions that can run on the processor, and when the processor runs the computer instructions, it executes the relevant content in any one of the above-mentioned molecular reverse synthesis methods based on quantum graph convolution.

[0035] It should be further noted that the technical features corresponding to the above options can be combined or replaced with each other without conflict to form a new technical solution.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] (1) The present invention represents the structural information and chemical properties of the target product in the form of a molecular undirected graph; maps each graph node and connection edge into a unique vector in the form of one-hot encoding; then performs quantum encoding on the molecular graph node and connection edge information data; constructs a quantum graph convolutional neural network based on the encoded quantum graph, outputs the predicted position of the reaction center to obtain a synthon graph; traverses the synthon graph using a graph convolutional policy network, identifies the atoms that need to be replaced or added and maps them into an action vector, and decodes the action vector to obtain the representation of the reactant. It can effectively depict the complex relationships between atoms in the molecule, effectively encode the molecular undirected graph into a quantum graph, realize the translation process from the product graph to the reactant graph, ensure the availability of the final reactant, the route prediction result is accurate, and the calculation difficulty is low, providing a technical reference for future drug research and development.

[0038] (2) Existing machine learning algorithms for retrosynthetic analysis are mainly reaction template-based methods. Although they have good interpretability, they require expensive subgraph matching, and once the matching fails, the model cannot give any predictions. The method proposed in the present invention is a method that does not rely on reaction templates, overcoming the limitation that R & D personnel need rich domain knowledge.

[0039] (3) Given the complexity and diversity of chemical reaction mechanisms, the reaction process includes a large number of intermediate products and transition states, resulting in extremely high complexity in predicting reaction centers. By virtue of the parallelism of quantum computing, multiple features of the product can be extracted simultaneously, which can accelerate the process of finding reaction centers, thereby shortening the time required to obtain accurate results and improving computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flowchart of a molecular reverse synthesis method based on quantum graph convolution shown in an embodiment of the present invention;

[0041] Figure 2 It is a schematic diagram of quantum state preparation shown in an embodiment of the present invention;

[0042] Figure 3 It is a schematic diagram of reaction center recognition based on quantum graph convolutional neural network shown in an embodiment of the present invention;

[0043] Figure 4 It is a schematic diagram of product-synthon transformation shown in an embodiment of the present invention;

[0044] Figure 5 It is a schematic diagram of a graph convolutional neural network model shown in an embodiment of the present invention;

[0045] Figure 6 It is a schematic diagram of a policy network model shown in an embodiment of the present invention;

[0046] Figure 7 It is an example diagram of synthon-reactant transformation shown in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0049] Refer to Figure 1, in an exemplary embodiment, a molecular retrosynthesis method based on quantum graph convolution is provided, including the following steps:

[0050] S1. Represent the structural information and chemical properties of the target product in the form of a molecular undirected graph, where the atoms of the target product are used as graph nodes and the chemical bonds between the atoms are used as connecting edges;

[0051] S2. Map each graph node and connecting edge into a unique vector in the way of one-hot encoding; then perform quantum encoding on the molecular graph node and edge information data and convert it into a corresponding quantum graph;

[0052] S3. Build a quantum graph convolutional neural network based on the encoded quantum graph, output the position of the predicted reaction center, disconnect the chemical bond where the reaction center is located, and then separate to obtain a synthon graph;

[0053] S4. Use a graph convolutional policy network to traverse the synthon graph, identify the atoms that need to be replaced or added, and map them into an action vector, and decode the action vector to obtain the representation of the reactants;

[0054] S5. Repeat steps S1 - S4, continuously update the reactants until the synthon graph is converted into the final reactants.

[0055] Specifically, use the method of molecular graph theory to represent the structural information and chemical properties of the target product in the form of a molecular undirected graph, with atoms as graph nodes and chemical bonds between atoms as connecting edges; map each node and edge into a unique vector in the way of one-hot encoding; then perform quantum encoding on the molecular graph node and edge information data and convert it into a corresponding quantum graph. For the quantum graph convolutional neural network, the input is the encoded quantum graph, and the output is the position of the predicted reaction center. By continuously stacking the quantum graph convolutional layer and the quantum pooling layer together, and finally adding the quantum fully connected layer and the output layer, a complete quantum graph convolutional neural network model is obtained. Predict the position of the reaction center according to the output vector of the quantum graph convolutional neural network, and then separate to obtain a synthon by disconnecting the chemical bond where the reaction center is located. Then, use the graph convolutional neural network to realize the feature extraction of nodes and edges and generate the feature vectors of nodes and edges, also called embedding vectors. Use the policy network to traverse the generated synthon graph, identify the atoms that need to be replaced or added, and map them into an action vector. Decoding the action vector can obtain the representation of the reactants. Finally, by repeatedly executing steps S1 - S4, continuously implement steps such as quantum state encoding of molecular graph data, reaction center identification, and synthon-reactant conversion, continuously update the reactants until the synthon graph is converted into the final reactants.

[0056] Among them, in the molecular retrosynthesis task, a database containing compound information is required to support research. Currently, there are also many compound databases available for reaction prediction, such as PubChem, ChEMBL, ZINC, DrugBank, ChemSpider, PDB, etc. These databases contain a large amount of compound information and can be used for tasks such as compound search, screening, and matching. When defining a molecular undirected graph, the molecular data obtained from the database is usually presented in the form of some structural information and chemical properties, such as the types and quantities of atoms in the molecule, the types and quantities of chemical bonds, the distances between atoms, etc. The molecular graph theory method is used to represent these structural information as an undirected graph.

[0057] In one example, a molecular retrosynthesis method based on quantum graph convolution, the quantum encoding of the molecular graph node and connection edge information data includes:

[0058] Using the angle encoding method to encode and convert the input vector into a superposition quantum state, and the conversion formula is as follows:

[0059]

[0060] Among them, θ represents the rotation angle of the quantum rotation gate, and x i represents the vector obtained by one-hot encoding.

[0061] Specifically, first, each node and edge are mapped to a unique vector using the one-hot encoding method, and then the molecular graph node and edge information data are quantum encoded and converted into a corresponding quantum graph. The classical data x obtained by one-hot encoding i and the rotation angle θ of the quantum rotation gate are constructed to have a corresponding relationship, so as to convert the characteristic information of the molecular undirected graph into the angle of the quantum rotation gate. Different quantum rotation gates R x (θ) act on the initial state |0> of the corresponding qubit, so as to retain the characteristic information in the quantum state and complete the preparation of the quantum state of the data. The schematic diagram of the preparation of the quantum state of the data is as shown in Figure 2 shown.

[0062] In one example, referring to Figure 3 , the quantum graph convolutional neural network includes a quantum graph convolutional layer, a quantum pooling layer, and a quantum fully connected layer. The input of the quantum graph convolutional neural network is the quantum graph obtained by encoding and processing the quantum state of the molecular graph data, and the output is the reaction center probability distribution vector, that is, the reactivity score.

[0063] Specifically, the step S3 specifically includes:

[0064] S31. Construct a quantum graph convolutional layer: Apply Hadamard gates and phase rotation gate operations to the input quantum circuit to perform a convolutional operation on the encoded quantum state, thereby extracting the features of the quantum graph and updating the state to achieve information transfer;

[0065] S32. Construct a quantum pooling layer: Sum each tensor output by the quantum graph convolution, use CNOT gates to control whether the summation result of each tensor is retained, and use SWAP gates to combine all the retained tensors, thereby obtaining the pooling result;

[0066] S33. Construct a quantum fully connected layer: Flatten the output of the quantum pooling layer into a vector, perform information processing through a quantum neural network layer, and output the prediction result of the reaction center;

[0067] S34. Construct a complete quantum graph convolutional neural network: Stack the quantum graph convolutional layer and the quantum pooling layer alternately, and then add a quantum fully connected layer and an output layer to form a complete quantum graph convolutional neural network model.

[0068] Further, referring to Figure 4 , the information processing through the quantum neural network layer and outputting the prediction result of the reaction center includes:

[0069] The quantum graph convolutional neural network outputs a reaction center probability distribution vector, that is, a reactivity score, selects the edge of the reaction center probability distribution vector with the highest score as the reaction center, and finally, separates the synthon by breaking the chemical bond where the reaction center is located.

[0070] In one example, a molecular retrosynthesis method based on quantum graph convolution, before traversing the synthon graph using the graph convolution policy network, further includes:

[0071] Use a graph convolutional neural network to extract the features of nodes and edges based on the input synthon graph and generate feature vectors of nodes and edges.

[0072] Specifically, referring to Figure 5 , the graph convolutional neural network includes an input layer, a hidden layer, and an output layer. The input layer inputs the synthon graph, and the hidden layer extracts the features of nodes and edges through convolution and a non-linear activation function and generates an embedding vector of nodes and edges.

[0073] Further, a molecular retrosynthesis method based on quantum graph convolution, the identifying atoms that need to be replaced or added and mapping them into an action vector includes:

[0074] Build a policy network. The policy network is a recurrent neural network based on LSTM, which can generate an action vector based on the embedding vector of the synthon graph for updating the synthon graph. The policy network model is asFigure 6 as shown

[0075] Furthermore, a graph convolutional neural network and a policy network are combined to form a graph convolutional policy network. The generated action vector is used to update the synthetic subgraph, thereby realizing the conversion of the synthetic subgraph to the reactant. The conversion process is as Figure 7 shown. The reactant generated by the graph convolutional policy network is used as the product again, and S1 - S4 are continuously repeated to update the reactant until the product is converted into the final reactant.

[0076] In another exemplary embodiment, a molecular retrosynthesis system based on quantum graph convolution is provided. The system includes:

[0077] A molecular undirected graph construction module configured to represent the structural information and chemical properties of a target product in the form of a molecular undirected graph, where the atoms of the target product are used as graph nodes and the chemical bonds between the atoms are used as connecting edges;

[0078] A quantum state encoding module configured to map each graph node and connecting edge to a unique vector in a one - hot encoding manner; and then perform quantum encoding on the molecular graph node and connecting edge information data to convert it into a corresponding quantum graph;

[0079] A reaction center recognition module configured to construct a quantum graph convolutional neural network based on the encoded quantum graph, output the position of the predicted reaction center, and break the chemical bond where the reaction center is located to separate and obtain a synthetic subgraph;

[0080] A synthetic - reactant conversion module configured to traverse the synthetic subgraph using a graph convolutional policy network, identify the atoms that need to be replaced or added and map them into an action vector. After repeating multiple times, the action vector is decoded to obtain the representation of the reactant. By repeatedly executing the above - mentioned several modules, the reactant is continuously updated until the synthetic subgraph is converted into the final reactant.

[0081] In another exemplary embodiment, the present invention provides a computer storage medium, on which computer instructions are stored, and when the computer instructions run, they execute the relevant content in the above - mentioned molecular retrosynthesis method based on quantum graph convolution.

[0082] Based on such an understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0083] In another exemplary embodiment, the present invention provides a terminal, including a memory and a processor. A computer instruction that can run on the processor is stored on the memory. When the processor runs the computer instruction, it executes the relevant content in the described method for molecular retrosynthesis based on quantum graph convolution.

[0084] The processor can be a single-core or multi-core central processing unit or a specific integrated circuit, or an integrated circuit configured to implement one or more of the present invention.

[0085] The embodiments of the subject matter and the functional operations described in this specification can be implemented in the following: tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more of them. The embodiments of the subject matter described in this specification can be implemented as one or more computer programs, that is, one or more modules in computer program instructions encoded on a tangible non-transitory program carrier to be executed by a data processing device or to control the operation of the data processing device. Alternatively or additionally, the program instructions can be encoded on an artificially generated propagated signal, such as a machine-generated electrical, optical, or electromagnetic signal, which is generated to encode and transmit information to a suitable receiver device for execution by the data processing device.

[0086] The processes and logical flows described in this specification can be executed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating on input data and generating outputs. The processes and logical flows can also be executed by dedicated logic circuits - such as FPGAs (field programmable gate arrays) or ASICs (application specific integrated circuits), and the device can also be implemented as dedicated logic circuits.

[0087] Processors suitable for executing computer programs include, for example, general and / or special purpose microprocessors, or any other type of central processing unit. Generally, the central processing unit will receive instructions and data from read-only memory and / or random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include one or more mass storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, etc., or the computer will be operably coupled to such mass storage devices to receive data therefrom or transfer data thereto, or both. However, a computer is not necessarily required to have such devices. In addition, a computer may be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name just a few.

[0088] Although this specification contains many specific implementation details, these should not be construed as limiting the scope of any invention or the scope of what is claimed, but rather as mainly describing the features of specific embodiments of a particular invention. Certain features described in multiple embodiments in this specification may also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination. In addition, although features may operate in certain combinations as described above and are even initially claimed as such, one or more features from a claimed combination may in some cases be removed from that combination, and the claimed combination may be directed to a sub-combination or a variation of a sub-combination.

[0089] Similarly, although operations are depicted in the drawings in a particular order, this should not be construed as requiring that the operations be performed in the particular order shown or sequentially, or that all illustrated operations be performed, to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of the various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0090] The above specific implementation manners are detailed descriptions of the present invention. It cannot be determined that the specific implementation manners of the present invention are only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions and substitutions can still be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A molecular retrosynthesis method based on quantum graph convolution, characterized in that, it includes the following steps: S1. Represent the structural information and chemical properties of the target product in the form of a molecular undirected graph, where the atoms of the target product are used as graph nodes, and the chemical bonds between the atoms are used as connecting edges; S2. Map each graph node and connecting edge into a unique vector in the way of one-hot encoding; then perform quantum encoding on the molecular graph node and connecting edge information data to convert it into a corresponding quantum graph; S3. Construct a quantum graph convolutional neural network based on the encoded quantum graph, output the position of the predicted reaction center, disconnect the chemical bond where the reaction center is located, and then separate to obtain a synthon graph; The step S3 specifically includes: S31. Apply Hadamard gate and phase rotation gate operations to the input quantum circuit to perform convolutional operations on the encoded quantum state, so as to extract the features of the quantum graph and update the state to achieve information transfer; S32. Sum each tensor output by the quantum graph convolution, use the CNOT gate to control whether the summation result of each tensor is retained, and use the SWAP gate to combine all the retained tensors to obtain a pooling result; S33. Flatten the output of the quantum pooling layer into a vector, perform information processing through the quantum neural network layer, and output the prediction result of the reaction center; The information processing through the quantum neural network layer and outputting the prediction result of the reaction center includes: The quantum graph convolutional neural network outputs a reaction center probability distribution vector, and selects the edge with the highest score of the reaction center probability distribution vector as the reaction center; S4. Use a graph convolutional policy network to traverse the synthon graph, identify the atoms that need to be replaced or added, and map them into an action vector, and decode the action vector to obtain the representation of the reactant; The graph convolutional policy network is formed by combining a graph convolutional neural network and a policy network, and the policy network is a recurrent neural network based on LSTM; The graph convolutional neural network extracts the features of nodes and edges according to the input synthon graph, and generates feature vectors of nodes and edges, that is, embedding vectors; The policy network generates an action vector according to the embedding vector of the synthon graph to update the synthon graph, and converts the updated synthon graph into a reactant; S5. Take the reactant generated in step S4 as the target product in step S1, repeat steps S1-S4, continuously update the reactant until the synthon graph is converted into the final reactant.

2. The molecular retrosynthesis method based on quantum graph convolution according to claim 1, characterized in that, the quantum encoding of the molecular graph node and connecting edge information data includes: Using the angle encoding method to encode and convert the input vector into a superposition quantum state, and the conversion formula is as follows: , where represents the rotation angle of the quantum rotation gate, represents the vector obtained by one-hot encoding.

3. The molecular retrosynthesis method based on quantum graph convolution according to claim 1, characterized in that, the quantum graph convolutional neural network includes a quantum graph convolutional layer, a quantum pooling layer and a quantum fully connected layer.

4. A molecular retrosynthesis system based on quantum graph convolution, characterized in that, the system includes: A molecular undirected graph construction module, configured to represent the structural information and chemical properties of a target product in the form of a molecular undirected graph, wherein the atoms of the target product are used as graph nodes, and the chemical bonds between the atoms are used as connecting edges; A quantum state encoding module, configured to map each graph node and connecting edge into a unique vector by using one-hot encoding; and then perform quantum encoding on the molecular graph node and connecting edge information data to convert it into a corresponding quantum graph; A reaction center recognition module, configured to construct a quantum graph convolutional neural network based on the encoded quantum graph, output the predicted position of the reaction center, and disconnect the chemical bond where the reaction center is located to separate the synthon graph; the constructing a quantum graph convolutional neural network based on the encoded quantum graph and outputting the predicted position of the reaction center includes: Applying Hadamard gate and phase rotation gate operations to the input quantum circuit to perform convolutional operations on the encoded quantum state, so as to extract the features of the quantum graph and update the state to achieve information transfer; Summing each tensor output by the quantum graph convolution, using the CNOT gate to control whether the summation result of each tensor is retained, and using the SWAP gate to combine all the retained tensors to obtain a pooling result; Flattening the output of the quantum pooling layer into a vector, performing information processing through a quantum neural network layer, and outputting the predicted result of the reaction center; The performing information processing through the quantum neural network layer and outputting the predicted result of the reaction center includes: The quantum graph convolutional neural network outputs a reaction center probability distribution vector, and selects the edge with the highest score in the reaction center probability distribution vector as the reaction center; A synthon-reactant conversion module, configured to traverse the synthon graph by using a graph convolutional policy network, identify the atoms that need to be replaced or added, and map them into an action vector. After repeating multiple times, decode the action vector to obtain the representation of the reactant; the graph convolutional policy network is formed by combining a graph convolutional neural network and a policy network, and the policy network is a recurrent neural network based on LSTM; The graph convolutional neural network extracts the features of nodes and edges according to the input synthon graph and generates feature vectors of nodes and edges, i.e., embedding vectors; The policy network generates an action vector according to the embedding vector of the synthon graph to update the synthon graph, and converts the updated synthon graph into a reactant; Continuously update the target product in the molecular undirected graph construction module by using the reactant generated in the synthon-reactant conversion module until the synthon graph is converted into the final reactant.

5. A computer storage medium, on which computer instructions are stored, characterized in that, when the computer instructions run, they execute the relevant content in the method for molecular retrosynthesis based on quantum graph convolution according to any one of claims 1-3.

6. A terminal, including a memory and a processor, and computer instructions that can run on the processor are stored on the memory, characterized in that, when the processor runs the computer instructions, it executes the relevant content in the method for molecular retrosynthesis based on quantum graph convolution according to any one of claims 1-3.

Images